blob: 62cbd3b8996162525756ba9280e652d13c191d68 [file] [log] [blame]
Stefan Hajnoczi75411d22010-12-06 16:08:00 +00001/*
2 * QEMU Enhanced Disk Format
3 *
4 * Copyright IBM, Corp. 2010
5 *
6 * Authors:
7 * Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
8 * Anthony Liguori <aliguori@us.ibm.com>
9 *
10 * This work is licensed under the terms of the GNU LGPL, version 2 or later.
11 * See the COPYING.LIB file in the top-level directory.
12 *
13 */
14
15#ifndef BLOCK_QED_H
16#define BLOCK_QED_H
17
18#include "block_int.h"
19
20/* The layout of a QED file is as follows:
21 *
22 * +--------+----------+----------+----------+-----+
23 * | header | L1 table | cluster0 | cluster1 | ... |
24 * +--------+----------+----------+----------+-----+
25 *
26 * There is a 2-level pagetable for cluster allocation:
27 *
28 * +----------+
29 * | L1 table |
30 * +----------+
31 * ,------' | '------.
32 * +----------+ | +----------+
33 * | L2 table | ... | L2 table |
34 * +----------+ +----------+
35 * ,------' | '------.
36 * +----------+ | +----------+
37 * | Data | ... | Data |
38 * +----------+ +----------+
39 *
40 * The L1 table is fixed size and always present. L2 tables are allocated on
41 * demand. The L1 table size determines the maximum possible image size; it
42 * can be influenced using the cluster_size and table_size values.
43 *
44 * All fields are little-endian on disk.
45 */
46
47enum {
48 QED_MAGIC = 'Q' | 'E' << 8 | 'D' << 16 | '\0' << 24,
49
50 /* The image supports a backing file */
51 QED_F_BACKING_FILE = 0x01,
52
Stefan Hajnoczi01979a92010-12-06 16:08:03 +000053 /* The image needs a consistency check before use */
54 QED_F_NEED_CHECK = 0x02,
55
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000056 /* The backing file format must not be probed, treat as raw image */
57 QED_F_BACKING_FORMAT_NO_PROBE = 0x04,
58
59 /* Feature bits must be used when the on-disk format changes */
60 QED_FEATURE_MASK = QED_F_BACKING_FILE | /* supported feature bits */
Stefan Hajnoczi01979a92010-12-06 16:08:03 +000061 QED_F_NEED_CHECK |
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000062 QED_F_BACKING_FORMAT_NO_PROBE,
63 QED_COMPAT_FEATURE_MASK = 0, /* supported compat feature bits */
64 QED_AUTOCLEAR_FEATURE_MASK = 0, /* supported autoclear feature bits */
65
66 /* Data is stored in groups of sectors called clusters. Cluster size must
67 * be large to avoid keeping too much metadata. I/O requests that have
68 * sub-cluster size will require read-modify-write.
69 */
70 QED_MIN_CLUSTER_SIZE = 4 * 1024, /* in bytes */
71 QED_MAX_CLUSTER_SIZE = 64 * 1024 * 1024,
72 QED_DEFAULT_CLUSTER_SIZE = 64 * 1024,
73
74 /* Allocated clusters are tracked using a 2-level pagetable. Table size is
75 * a multiple of clusters so large maximum image sizes can be supported
76 * without jacking up the cluster size too much.
77 */
78 QED_MIN_TABLE_SIZE = 1, /* in clusters */
79 QED_MAX_TABLE_SIZE = 16,
80 QED_DEFAULT_TABLE_SIZE = 4,
Stefan Hajnoczi6f321e92011-05-09 16:45:40 +010081
82 /* Delay to flush and clean image after last allocating write completes */
83 QED_NEED_CHECK_TIMEOUT = 5, /* in seconds */
Stefan Hajnoczi75411d22010-12-06 16:08:00 +000084};
85
86typedef struct {
87 uint32_t magic; /* QED\0 */
88
89 uint32_t cluster_size; /* in bytes */
90 uint32_t table_size; /* for L1 and L2 tables, in clusters */
91 uint32_t header_size; /* in clusters */
92
93 uint64_t features; /* format feature bits */
94 uint64_t compat_features; /* compatible feature bits */
95 uint64_t autoclear_features; /* self-resetting feature bits */
96
97 uint64_t l1_table_offset; /* in bytes */
98 uint64_t image_size; /* total logical image size, in bytes */
99
100 /* if (features & QED_F_BACKING_FILE) */
101 uint32_t backing_filename_offset; /* in bytes from start of header */
102 uint32_t backing_filename_size; /* in bytes */
103} QEDHeader;
104
105typedef struct {
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000106 uint64_t offsets[0]; /* in bytes */
107} QEDTable;
108
109/* The L2 cache is a simple write-through cache for L2 structures */
110typedef struct CachedL2Table {
111 QEDTable *table;
112 uint64_t offset; /* offset=0 indicates an invalidate entry */
113 QTAILQ_ENTRY(CachedL2Table) node;
114 int ref;
115} CachedL2Table;
116
117typedef struct {
118 QTAILQ_HEAD(, CachedL2Table) entries;
119 unsigned int n_entries;
120} L2TableCache;
121
122typedef struct QEDRequest {
123 CachedL2Table *l2_table;
124} QEDRequest;
125
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000126typedef struct QEDAIOCB {
127 BlockDriverAIOCB common;
128 QEMUBH *bh;
129 int bh_ret; /* final return status for completion bh */
130 QSIMPLEQ_ENTRY(QEDAIOCB) next; /* next request */
131 bool is_write; /* false - read, true - write */
132 bool *finished; /* signal for cancel completion */
133 uint64_t end_pos; /* request end on block device, in bytes */
134
135 /* User scatter-gather list */
136 QEMUIOVector *qiov;
137 size_t qiov_offset; /* byte count already processed */
138
139 /* Current cluster scatter-gather list */
140 QEMUIOVector cur_qiov;
141 uint64_t cur_pos; /* position on block device, in bytes */
142 uint64_t cur_cluster; /* cluster offset in image file */
143 unsigned int cur_nclusters; /* number of clusters being accessed */
144 int find_cluster_ret; /* used for L1/L2 update */
145
146 QEDRequest request;
147} QEDAIOCB;
148
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000149typedef struct {
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000150 BlockDriverState *bs; /* device */
151 uint64_t file_size; /* length of image file, in bytes */
152
153 QEDHeader header; /* always cpu-endian */
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000154 QEDTable *l1_table;
155 L2TableCache l2_cache; /* l2 table cache */
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000156 uint32_t table_nelems;
157 uint32_t l1_shift;
158 uint32_t l2_shift;
159 uint32_t l2_mask;
Stefan Hajnoczieabba582010-12-06 16:08:02 +0000160
161 /* Allocating write request queue */
162 QSIMPLEQ_HEAD(, QEDAIOCB) allocating_write_reqs;
Stefan Hajnoczi6f321e92011-05-09 16:45:40 +0100163 bool allocating_write_reqs_plugged;
164
165 /* Periodic flush and clear need check flag */
166 QEMUTimer *need_check_timer;
Anthony Liguori1ed520c2011-11-14 15:09:47 -0600167
168 Error *migration_blocker;
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000169} BDRVQEDState;
170
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000171enum {
172 QED_CLUSTER_FOUND, /* cluster found */
Anthony Liguori21df65b2010-12-17 15:58:22 +0000173 QED_CLUSTER_ZERO, /* zero cluster found */
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000174 QED_CLUSTER_L2, /* cluster missing in L2 */
175 QED_CLUSTER_L1, /* cluster missing in L1 */
176};
177
178/**
179 * qed_find_cluster() completion callback
180 *
181 * @opaque: User data for completion callback
182 * @ret: QED_CLUSTER_FOUND Success
183 * QED_CLUSTER_L2 Data cluster unallocated in L2
184 * QED_CLUSTER_L1 L2 unallocated in L1
185 * -errno POSIX error occurred
186 * @offset: Data cluster offset
187 * @len: Contiguous bytes starting from cluster offset
188 *
189 * This function is invoked when qed_find_cluster() completes.
190 *
191 * On success ret is QED_CLUSTER_FOUND and offset/len are a contiguous range
192 * in the image file.
193 *
194 * On failure ret is QED_CLUSTER_L2 or QED_CLUSTER_L1 for missing L2 or L1
195 * table offset, respectively. len is number of contiguous unallocated bytes.
196 */
197typedef void QEDFindClusterFunc(void *opaque, int ret, uint64_t offset, size_t len);
198
199/**
200 * Generic callback for chaining async callbacks
201 */
202typedef struct {
203 BlockDriverCompletionFunc *cb;
204 void *opaque;
205} GenericCB;
206
207void *gencb_alloc(size_t len, BlockDriverCompletionFunc *cb, void *opaque);
208void gencb_complete(void *opaque, int ret);
209
210/**
211 * L2 cache functions
212 */
213void qed_init_l2_cache(L2TableCache *l2_cache);
214void qed_free_l2_cache(L2TableCache *l2_cache);
215CachedL2Table *qed_alloc_l2_cache_entry(L2TableCache *l2_cache);
216void qed_unref_l2_cache_entry(CachedL2Table *entry);
217CachedL2Table *qed_find_l2_cache_entry(L2TableCache *l2_cache, uint64_t offset);
218void qed_commit_l2_cache_entry(L2TableCache *l2_cache, CachedL2Table *l2_table);
219
220/**
221 * Table I/O functions
222 */
223int qed_read_l1_table_sync(BDRVQEDState *s);
224void qed_write_l1_table(BDRVQEDState *s, unsigned int index, unsigned int n,
225 BlockDriverCompletionFunc *cb, void *opaque);
226int qed_write_l1_table_sync(BDRVQEDState *s, unsigned int index,
227 unsigned int n);
228int qed_read_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
229 uint64_t offset);
230void qed_read_l2_table(BDRVQEDState *s, QEDRequest *request, uint64_t offset,
231 BlockDriverCompletionFunc *cb, void *opaque);
232void qed_write_l2_table(BDRVQEDState *s, QEDRequest *request,
233 unsigned int index, unsigned int n, bool flush,
234 BlockDriverCompletionFunc *cb, void *opaque);
235int qed_write_l2_table_sync(BDRVQEDState *s, QEDRequest *request,
236 unsigned int index, unsigned int n, bool flush);
237
238/**
239 * Cluster functions
240 */
241void qed_find_cluster(BDRVQEDState *s, QEDRequest *request, uint64_t pos,
242 size_t len, QEDFindClusterFunc *cb, void *opaque);
243
244/**
245 * Consistency check
246 */
247int qed_check(BDRVQEDState *s, BdrvCheckResult *result, bool fix);
248
249QEDTable *qed_alloc_table(BDRVQEDState *s);
250
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000251/**
252 * Round down to the start of a cluster
253 */
254static inline uint64_t qed_start_of_cluster(BDRVQEDState *s, uint64_t offset)
255{
256 return offset & ~(uint64_t)(s->header.cluster_size - 1);
257}
258
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000259static inline uint64_t qed_offset_into_cluster(BDRVQEDState *s, uint64_t offset)
260{
261 return offset & (s->header.cluster_size - 1);
262}
263
Stefan Hajnoczi19dfc442011-04-24 18:38:58 +0100264static inline uint64_t qed_bytes_to_clusters(BDRVQEDState *s, uint64_t bytes)
Stefan Hajnoczi298800c2010-12-06 16:08:01 +0000265{
266 return qed_start_of_cluster(s, bytes + (s->header.cluster_size - 1)) /
267 (s->header.cluster_size - 1);
268}
269
270static inline unsigned int qed_l1_index(BDRVQEDState *s, uint64_t pos)
271{
272 return pos >> s->l1_shift;
273}
274
275static inline unsigned int qed_l2_index(BDRVQEDState *s, uint64_t pos)
276{
277 return (pos >> s->l2_shift) & s->l2_mask;
278}
279
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000280/**
281 * Test if a cluster offset is valid
282 */
283static inline bool qed_check_cluster_offset(BDRVQEDState *s, uint64_t offset)
284{
285 uint64_t header_size = (uint64_t)s->header.header_size *
286 s->header.cluster_size;
287
288 if (offset & (s->header.cluster_size - 1)) {
289 return false;
290 }
291 return offset >= header_size && offset < s->file_size;
292}
293
294/**
295 * Test if a table offset is valid
296 */
297static inline bool qed_check_table_offset(BDRVQEDState *s, uint64_t offset)
298{
299 uint64_t end_offset = offset + (s->header.table_size - 1) *
300 s->header.cluster_size;
301
302 /* Overflow check */
303 if (end_offset <= offset) {
304 return false;
305 }
306
307 return qed_check_cluster_offset(s, offset) &&
308 qed_check_cluster_offset(s, end_offset);
309}
310
Anthony Liguori21df65b2010-12-17 15:58:22 +0000311static inline bool qed_offset_is_cluster_aligned(BDRVQEDState *s,
312 uint64_t offset)
313{
314 if (qed_offset_into_cluster(s, offset)) {
315 return false;
316 }
317 return true;
318}
319
320static inline bool qed_offset_is_unalloc_cluster(uint64_t offset)
321{
322 if (offset == 0) {
323 return true;
324 }
325 return false;
326}
327
328static inline bool qed_offset_is_zero_cluster(uint64_t offset)
329{
330 if (offset == 1) {
331 return true;
332 }
333 return false;
334}
335
Stefan Hajnoczi75411d22010-12-06 16:08:00 +0000336#endif /* BLOCK_QED_H */